Methodology

How the numbers get made.

How the score works

Each city's score is a composite index built from nineteen weighted inputs, grouped into three buckets with bucket caps, plus a small user-vote rider. Every metric is normalized to a 0..1 range; the buckets are combined under their caps and rescaled to 0..100, then a uniform 15% display boost is applied (capped at 100) so the visible distribution sits in a readable band instead of clustering mid-range. The same weight vector is applied to three views: a Native score (the host country), a Residentsscore (weighted by the origin composition of the city's female visitor and resident population), and a Combined score that blends the two 50/50. Current weight set: v6.

Inputs

Bucket A — Revealed international preference

cap 45%

  • Pageants

    30%

    Big Four international beauty pageant wins (Miss Universe, World, International, Earth), normalized per capita.

  • Modeling

    22%

    Share of internationally represented fashion models (Models.com Top 50, Vogue covers) per capita.

  • Supermodel rosters

    15%

    Representation at IMG / Elite / Ford / Wilhelmina / Storm agency rosters.

  • Dating apps

    15%

    Cross-border match data from published dating-app research.

  • K-1 visa

    8%

    US State Department K-1 fiancée visa issuance per capita.

  • Adult industry

    10%

    Country-of-origin performer prevalence (Pornhub annual data).

Bucket B — Physical / biological measurement

cap 30%

  • BMI fit

    22%

    Proximity of national female BMI to the 19–24 band (WHO).

  • WHR fit

    22%

    Proximity of national female WHR to 0.70 (DHS).

  • Obesity (inverse)

    18%

    Inverse of female obesity rate (WHO).

  • Symmetry proxy

    13%

    Genetic heterozygosity / outbreeding coefficient.

  • Height (z-score)

    13%

    National female average height (NCD-RisC).

  • Skin health

    12%

    Inverse composite of UV exposure and PM2.5 (WHO / satellite).

Bucket C — Cultural consensus

cap 25%

  • Consensus (academic)

    22%

    Aggregated results from published cross-cultural rater studies (Czech faces dataset, PLOS ONE, Chicago Face Database).

  • Consensus (forums)

    18%

    Aggregated sentiment from travel, expat, and dating communities online.

  • Instagram density

    13%

    Geo-tagged beauty/fashion content density per capita.

  • Cosmetic spend

    10%

    Per-capita cosmetics industry spend (Euromonitor).

  • CV facial regressor

    17%

    Country-level means from published CV facial-attractiveness regressors (SCUT-FBP5500, He et al. 2023), ancestry-cluster mapped. Raw outputs — reflects training-data representation, not corrected.

  • Mention density

    10%

    Log-normalized frequency of mentions across vote comments and posts over a rolling 90-day window. Crowd-sourced consensus signal derived from in-app content; feeds back into the entity's cultural score.

  • Venue score

    6%

    Aggregated community score of approved venues belonging to the entity (cities) or its constituent cities (countries). Empty or low-sample entities land at zero.

  • Beauty venue density

    4%

    Per-capita density of nail salons, lash/brow studios, blowout bars, med spas, and pilates/barre studios within a city radius. Blended from OpenStreetMap and Yelp Fusion: each category sqrt-rescaled to 0..1 across cities, then max(OSM, Yelp) per category. Captures commercial beauty infrastructure — distinct from the cultural/audience signals above. Tilts toward developed-market metros; sparse in cities where Yelp doesn't operate.

User votes

5%

Ratings submitted in-app. Low weight and Bayesian-smoothed so small samples don't swing rankings. Rides on top of the three bucket caps rather than competing inside them.

Access correction (Bucket A)

Bucket A metrics depend on things orthogonal to beauty: GDP, diaspora size in the US/EU, and domestic fashion/pageant infrastructure. Countries lacking all three can't show up in these metrics — absence of signal is not absence of quality. We fit an OLS model per metric against log(GDP per capita), diaspora intensity, and an industry-access index, and apply a one-sided lift: countries scoring below their predicted value are nudged up toward it; countries at or above prediction are untouched. Top performers stay put, under-represented countries drift up.

Native, Residents, Combined

Native
The score for the host country's own population, using the country-level metrics with any city-level overrides merged on top. This isolates how the locals rate on the index, ignoring who else happens to be in the city.
Residents
A presence-weighted score that accounts for the origin-country composition of the city's female visitor and resident population via the inflows field on each city. Each origin contributes its own country score, weighted by its share of the relevant population.
Combined
The default view. A 50/50 blend of Native and Residents. Balances who is from the city against who is in the city at any given time.

Confidence

Each score carries an associated confidence derived from the underlying sample size across all inputs. In the stats and leaderboard views, low-confidence entries are dimmed via reduced opacity so ranking and certainty can be read at a glance.

Sources

Versioning

Weights version
v6
Last updated
2026-05-05

Weights and inputs may change over time as new data and studies are incorporated. Scores are recomputed from the metric vector on every data import.